TB kills an estimated 1.23 million people a year, and a key bottleneck is the shortage of radiologists to read X-rays in remote clinics and mines. Google first showed in 2019 that deep learning could match experienced radiologists at spotting TB. Nexus AI’s device holds EU Class IIb certification, and the company is pursuing WHO prequalification and US FDA clearance. Indonesia aims to screen 50 million people a year and keeps patient data inside the country by running the models locally. For places without X-ray machines, Google’s HeAR model screens from cough sounds recorded on a smartphone.
Why it matters
This is a rare case of medical AI crossing the gap from published paper to certified bedside tool, a step that took audits and partners, not just a better model. Open models matter too: they let countries build screening on their own servers instead of sending patient scans abroad. Whether faster screening turns into more people treated is the open question.
Original source
Bringing foundational models to the front lines of TB screening
Google Research
research.google
Summary by Second Order AI. Full reporting is at the original source.